Si-Wei Liu
2022-Jan-07 01:25 UTC
[PATCH v7 05/14] vdpa: Allow to configure max data virtqueues
On 1/5/2022 3:46 AM, Eli Cohen wrote:> Add netlink support to configure the max virtqueue pairs for a device. > At least one pair is required. The maximum is dictated by the device. > > Example: > $ vdpa dev add name vdpa-a mgmtdev auxiliary/mlx5_core.sf.1 max_vqp 4 > > Signed-off-by: Eli Cohen <elic at nvidia.com> > --- > v6->v7: > 1.Serialize set_features and reset using cf_mutex to ensure consistency > with netlink set/get > > drivers/vdpa/vdpa.c | 15 +++++++++++++-- > drivers/vhost/vdpa.c | 2 +- > drivers/virtio/virtio_vdpa.c | 2 +- > include/linux/vdpa.h | 19 ++++++++++++++++--- > 4 files changed, 31 insertions(+), 7 deletions(-) > > diff --git a/drivers/vdpa/vdpa.c b/drivers/vdpa/vdpa.c > index 4494325cae91..96d31b80fdce 100644 > --- a/drivers/vdpa/vdpa.c > +++ b/drivers/vdpa/vdpa.c > @@ -404,7 +404,7 @@ static void vdpa_get_config_unlocked(struct vdpa_device *vdev, > * If it does happen we assume a legacy guest. > */ > if (!vdev->features_valid) > - vdpa_set_features(vdev, 0); > + vdpa_set_features(vdev, 0, true);Can we do it here with an internal unlocked version vdpa_set_features_unlocked() without taking the cf_mutex? Such that all API users for vdpa_set_features() won't have to change the prototype. It looks to me the only place that needs the unlocked API is the vdpa core itself, which doesn't need to expose the internal API to other modules, is my understanding correct? In addition, it seems more appropriate to move the vdpa_set_features() related changes to a separate patch like patch #3. It's not obvious to me how it's logically connected to the code change for the max_vqp feature here (if there's anything it may be more relevant to patch #8 of this series).> ops->get_config(vdev, offset, buf, len); > } > > @@ -581,7 +581,8 @@ vdpa_nl_cmd_mgmtdev_get_dumpit(struct sk_buff *msg, struct netlink_callback *cb) > } > > #define VDPA_DEV_NET_ATTRS_MASK ((1 << VDPA_ATTR_DEV_NET_CFG_MACADDR) | \ > - (1 << VDPA_ATTR_DEV_NET_CFG_MTU)) > + (1 << VDPA_ATTR_DEV_NET_CFG_MTU) | \ > + (1 << VDPA_ATTR_DEV_NET_CFG_MAX_VQP)) > > static int vdpa_nl_cmd_dev_add_set_doit(struct sk_buff *skb, struct genl_info *info) > { > @@ -607,6 +608,16 @@ static int vdpa_nl_cmd_dev_add_set_doit(struct sk_buff *skb, struct genl_info *i > nla_get_u16(nl_attrs[VDPA_ATTR_DEV_NET_CFG_MTU]); > config.mask |= (1 << VDPA_ATTR_DEV_NET_CFG_MTU); > } > + if (nl_attrs[VDPA_ATTR_DEV_NET_CFG_MAX_VQP]) { > + config.net.max_vq_pairs > + nla_get_u16(nl_attrs[VDPA_ATTR_DEV_NET_CFG_MAX_VQP]); > + if (!config.net.max_vq_pairs) { > + NL_SET_ERR_MSG_MOD(info->extack, > + "At least one pair of VQs is required"); > + return -EINVAL; > + } > + config.mask |= BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MAX_VQP); > + } > > /* Skip checking capability if user didn't prefer to configure any > * device networking attributes. It is likely that user might have used > diff --git a/drivers/vhost/vdpa.c b/drivers/vhost/vdpa.c > index d9d499465e2e..c37a63ba620a 100644 > --- a/drivers/vhost/vdpa.c > +++ b/drivers/vhost/vdpa.c > @@ -285,7 +285,7 @@ static long vhost_vdpa_set_features(struct vhost_vdpa *v, u64 __user *featurep) > if (copy_from_user(&features, featurep, sizeof(features))) > return -EFAULT; > > - if (vdpa_set_features(vdpa, features)) > + if (vdpa_set_features(vdpa, features, false)) > return -EINVAL; > > return 0; > diff --git a/drivers/virtio/virtio_vdpa.c b/drivers/virtio/virtio_vdpa.c > index 76504559bc25..7767a7f0119b 100644 > --- a/drivers/virtio/virtio_vdpa.c > +++ b/drivers/virtio/virtio_vdpa.c > @@ -317,7 +317,7 @@ static int virtio_vdpa_finalize_features(struct virtio_device *vdev) > /* Give virtio_ring a chance to accept features. */ > vring_transport_features(vdev); > > - return vdpa_set_features(vdpa, vdev->features); > + return vdpa_set_features(vdpa, vdev->features, false); > } > > static const char *virtio_vdpa_bus_name(struct virtio_device *vdev) > diff --git a/include/linux/vdpa.h b/include/linux/vdpa.h > index ae047fae2603..6d4d7e4fe208 100644 > --- a/include/linux/vdpa.h > +++ b/include/linux/vdpa.h > @@ -101,6 +101,7 @@ struct vdpa_dev_set_config { > struct { > u8 mac[ETH_ALEN]; > u16 mtu; > + u16 max_vq_pairs; > } net; > u64 mask; > }; > @@ -391,17 +392,29 @@ static inline struct device *vdpa_get_dma_dev(struct vdpa_device *vdev) > static inline int vdpa_reset(struct vdpa_device *vdev) > { > const struct vdpa_config_ops *ops = vdev->config; > + int ret; > > + mutex_lock(&vdev->cf_mutex); > vdev->features_valid = false; > - return ops->reset(vdev); > + ret = ops->reset(vdev); > + mutex_unlock(&vdev->cf_mutex); > + return ret; > }Can we move the vdpa_reset() code change here to patch #3? i.e. to be in parallel with set_status() changes. -Siwei> > -static inline int vdpa_set_features(struct vdpa_device *vdev, u64 features) > +static inline int vdpa_set_features(struct vdpa_device *vdev, u64 features, bool locked) > { > const struct vdpa_config_ops *ops = vdev->config; > + int ret; > + > + if (!locked) > + mutex_lock(&vdev->cf_mutex); > > vdev->features_valid = true; > - return ops->set_driver_features(vdev, features); > + ret = ops->set_driver_features(vdev, features); > + if (!locked) > + mutex_unlock(&vdev->cf_mutex); > + > + return ret; > } > > void vdpa_get_config(struct vdpa_device *vdev, unsigned int offset,